Learning acoustic word embeddings with phonetically associated triplet network
This work addresses the need for more discriminative embeddings in speech processing, offering an incremental enhancement over existing triplet network methods.
The paper tackled the problem of improving acoustic word embeddings for query-by-example spoken term detection by proposing a phonetically associated triplet network (PATN) that incorporates phonetic information, resulting in over 20% relative improvement in recall rate on the WSJ dataset.
Previous researches on acoustic word embeddings used in query-by-example spoken term detection have shown remarkable performance improvements when using a triplet network. However, the triplet network is trained using only a limited information about acoustic similarity between words. In this paper, we propose a novel architecture, phonetically associated triplet network (PATN), which aims at increasing discriminative power of acoustic word embeddings by utilizing phonetic information as well as word identity. The proposed model is learned to minimize a combined loss function that was made by introducing a cross entropy loss to the lower layer of LSTM-based triplet network. We observed that the proposed method performs significantly better than the baseline triplet network on a word discrimination task with the WSJ dataset resulting in over 20% relative improvement in recall rate at 1.0 false alarm per hour. Finally, we examined the generalization ability by conducting the out-of-domain test on the RM dataset.